DEV Community

Cover image for Artificial intelligence found a security flaw in medical software in eleven minutes
Hacks.gr
Hacks.gr

Posted on Originally published at en.hacks.gr Fully Autonomous

Artificial intelligence found a security flaw in medical software in eleven minutes

Amazon Threat Intelligence details an intrusion where suspected China-linked hackers used autonomous AI agents via the Cairn framework to find and exploit an unknown vulnerability in MONAI Label 0.8.5 and earlier.

The agents successfully mapped the interface and reached file-writing capabilities on unauthenticated servers within minutes.


What software is affected?

The issue concerns MONAI Label, an open-source program used to process medical scans and train models. In its default setting, the software does not ask for a login password, allowing anyone with network access to send files to locations outside of the designated folder.

How artificial intelligence worked

The case was uncovered when researchers discovered a control panel for Cairn, a program that allows artificial intelligence systems to operate autonomously, open online and without a password. The human operator set the goal of gaining access and selected the targets, while the programs chose the methods.

Within about eleven minutes, the algorithms managed to map the environment, find the weakness, and get to the point of writing files.

In a next step, the attackers exploited the issue to place a file that automatically runs when the Python programming language is opened. They then triggered a system training mode, which activated the file and gave them full administrator access.

↳

READ ALSO

HACKS.GR / RELATED

Cyber ​​Attacks & Data Breaches

Fake GTA VI app promises PC gameplay, but could steal data

↗

The perpetrators later gave the programs access to a second university network running the same software, although the actual breach of the second network remains unconfirmed.

The identity of the perpetrators and the correction

Researchers attribute the attack to a group from China with a moderate degree of certainty, due to instructions in simplified Chinese, time settings, and the selected targets.

The program's maintainers added a code fix on October 19, 2026, which forces uploaded files to be saved only in approved folders. However, a new ready-made installation package has not yet been released, and version 0.8.5 remains the latest official list.

Recommended protection measures include blocking access from untrusted networks with firewalls, enabling login passwords, using a limited account, and checking for any unusual Python startup files.

>Stay connected with Hacks.gr

Set us as a preferred source on Google and find us on social media.

Preferred source

↗

end of article


Read the original English article on Hacks.gr

Top comments (0)